Low Updating Frequency Model-Free Predictive Control Using Time-Series Model on PMSM Drives

The model-free predictive control (MFPC) strategy is commonly used to address the problem of weak robustness in model-based predictive control (MBPC). However, due to the estimation algorithm requiring large processor resources in this strategy, the possibility of overrun errors cannot be avoided fully. To release this problem, a low updating frequency MFPC is presented in this paper to prevent overrun error and reduce the influences on model accuracy. The algorithm is based on a time-series model with a recursive least square (RLS) estimation algorithm, and uses spline interpolation to obtain multiple groups of extended data and insert them into the regressive vector. The latest terms of the vector are used to generate coefficients in the model and predict the output signals to the next sampling period through multi-step prediction. The error and stability are analyzed theoretically, and the advantages of this method, including better model quality and current quality with enhanced robustness, are verified through experimental results.